Gradient boosting is the model that wins on tabular data, and XGBoost and LightGBM are the two implementations you will be expected to know by name in a quant interview. Where a random forest reduces variance by averaging independent trees in parallel, boosting reduces bias by building trees in sequence, each one correcting the errors of the ensemble so far. This lesson explains the core idea of fitting the residuals, shows why it is literally gradient descent in function space, works two boosting iterations by hand, and covers the regularization knobs that keep a powerful model from overfitting. We tie it to the bias-variance tradeoff, to the random forest comparison, and to how boosting shows up in trading.
Table of Contents
Already have an account? Log in!